Speaker Caption Merger: combine per-speaker SRT/VTT locally while preserving overlaps

[mozzie49]

I built Speaker Caption Merger, a small free, MIT-licensed browser tool for the step after transcribing separate speaker tracks with Whisper or Buzz. It combines existing timed subtitle files into one speaker-labeled conversation. This is my project, developed with AI assistance.

Live tool · Source and documentation

1. Add one UTF-8 SRT or VTT file per speaker; mixed formats work.
2. Set the speaker labels and, if needed, a constant time offset for each file.
3. Download labeled SRT, readable TXT/Markdown, or a JSON archive with the source text and cue data.

The “Try an example” button loads two synthetic files, Alex.srt and 小林.vtt. One cue runs from 1.000–3.500 seconds and the other from 2.800–4.600 seconds: both intervals remain intact in the merged output. Cues are sorted by start time without clipping overlapping speech or removing repeated text.

Parsing and merging happen locally in the browser tab, with no account, API key, or caption uploads. You can also download the repository and open index.html offline.

Scope: this is an independent post-export utility. It does no audio splitting/transcription, diarization, automatic synchronization, clock-drift correction, or alignment of untimed TXT. Timing is only as accurate as your inputs and chosen offsets. Some subtitle players hide or replace overlapping SRT cues; TXT/Markdown keeps both speakers’ timed entries available for review. VTT styling/positioning is kept in JSON but omitted from the transcript/SRT exports.

If you try it with separate-speaker Whisper exports, does the labeled SRT keep the expected timings when imported into your usual subtitle editor or player? The application name and a small non-private example of any mismatch would help.